This analysis and report was undertaken purely as an academic exercise. The assumptions in this report are fundamentally different and inconsistent with the framework of the Philadelphia Water Department Consent Order & Agreement. The results and conclusions of the report are not to be used to infer or inform compliance with regulatory requirements.

Abstract

The purpose of this study is to determine the degree to which the Philadelphia Water Department (PWD) Green City Clean Waters program has affected combined sewer overflows (CSO) into surrounding waterways and determine factors associated with a reduction in overflows. The study uses rainfall data from 2009-2017 to calibrate reported CSO overflow metrics from 2009-2016. The study compares individual CSO outfall volume and frequency to land cover, land use, stormwater management practices (SMP) constructed prior to 2017, as well as PWD’s various stormwater infrastructure programs (redevelopment regulations, private retrofit incentives, street projects, and public parcel projects).

The results show a slight overall reduction in the volume and frequency of overflows throughout the combined sewer system (CSS). The results also indicate that some CSO drainage areas (sewersheds) have seen higher reductions than others and that certain programs are associated with higher overflow reductions. Additionally, the data also show that there are sewersheds that have a higher amount of stormwater investment relative to overflow volume and there are sewersheds where current investment is not commensurate with system performance needs. The intent of the investigation is to identify performance-based factors that can help guide future stormwater infrastructure investment in Philadelphia.

Introduction

Since the Philadelphia Water Department’s (PWD) Green City, Clean Waters program was started in 2011, more than 1,160 greened acres (acre-inches) of green stormwater infrastructure capacity has been constructed throughout the Combined Sewer Service area (CSS). Theoretically this equates to a system-wide reduction of 31.5 million gallons of CSO overflow for most heavy rain events. Under the Consent Order & Agreement, PWD is mandated to construct over the 25-year program 9,564 acre-inches of stormwater capacity and a total annual overflow volume reduction of at least 7.96 billion gallons. The agreement specifies various benchmarks that PWD must hit at various increments. At year 5, PWD exceeded its 5-year Greened Acre benchmark as well as the Overflow Reduction Volume benchmark. However, the Agreement does not specify where green infrastructure must be located nor which CSO volumes must be reduced, but rather the benchmarks are set for the entire CSS. However, as the benchmark values increase and the initial round of projects has been constructed, it will likely become more challenging to plan for new infrastructure projects.

The purpose of this study is to measure which CSOs have seen the highest volume reductions and understand with more nuance what factors are associated with their higher performance in hopes of providing insight that could help PWD to plan as much for the Overflow Volume Reduction benchmarks as it does for the Greened Acre benchmarks in order for Green City, Clean Waters to continue to hit its milestones.

Methods

The data collected for this study were all from publicly available sources except for the sewershed boundary shapefile, which was provided by PWD. Given the sensitive nature of the sewersheds boundaries, all direct representations of the sewersheds in this report have been removed. The sewersheds were used to derive the dataset and then removed from the dataset. For illustrative purposes, this report visually represents sewersheds using computed watersheds that were processed from a publicly available Digital Elevation Model using the hydrological modeling program ArcHydro. While some of the sewershed boundaries are quite similar to the natural watershed boundaries, the maps contained herein do not represent PWD sewershed boundaries.

The methods section is a brief explanation of the data wrangling process. More detail for each of the data sources is included in the Appendix. Links to corresponding data sources and references to the appendix are included throughout in order to provide finer-grain detail for those who are interested.

Definitions

Term Definition
Greened Acre Volume of 1 acre-inch
Overflow Percent The total volume of annual CSO overflow divided the total annual volume of precipitaition the sewershed recieved
Overflow Volume Intensity Total annual CSO overflow volume (acre-inches) divided by total number of overflow events
Precipitation Volume Intensity Total annual precipitation volume (acre-inches) divided by total annual precipitation events
Rate Change The rate change in percent overflow over the study period (the slope of the linear regression fit line with percent overflow as dependent variable)
Sewershed Drainage area of specified CSO regulator
SMP Stormwater management practice (type of stormwater infrastructure)

Precipitation

In order to assess the variation in precipitation from year to year, daily precipitation summaries from July 1, 2008 through June 30, 2018 from 5 NOAA weather stations throughout the Philadelphia metro region was gathered via the NOAA API. The daily totals for each of the weather stations show variations in storm intensity across the Philadelphia region. The daily precipitation values were aggregated to annual summaries based on the reporting period for the corresponding NPDES stormwater management program annual report (July 1 - June 31). This allows for a direct comparison between precipitation and overflows for each CSO.

Given the relatively wide variation in annual precipitation metrics across the 5 weather stations, it was necessary to blend the values in order to obtain averaged values. Inverse Distance Weighting (IDW) was used to impute raster surfaces for each precipitation metric for each year. IDW allows for a geographic blending of the various weather station data so that coordinates inherit average values of the nearest weather stations weighted by proximity to each station. The total precipitation rasters were extracted by the sewershed boundaries in order to obtain the total volume of precipitation that fell within each sewershed boundary for each study year. The total precipitation volume allows for a direct comparison between the volume of precipitation and the annual volume (in acre-inches) overflow for each CSO regulator.

CSO Points & Sewersheds

All data in this report were aggregated by the corresponding sewershed boundaries using the sewersheds shapefile obtained from PWD. All annual CSO overflow metrics (volume, duration, frequency) were obtained from the NPDES stormwater management program annual reports from 2008 through 2018. CSO outfalls were geolocated using the coordinates reported in the listing of permitted CSO outfalls table of the NPDES reports.

Overflow volume, frequency, and duration reported for fiscal year 2018 is identical to that of fiscal year 2017, despite the increased precipitation in fiscal year 2018. Therefore, precipitation data from fiscal year 2018 and stormwater projects completed in fiscal year 2017 have been removed from the dataset since there is no way of measuring their effects on overflow in the following year.

Given the interconnected nature of Philadelphia’s sewer system, it is not possible to compare one-for-one all CSO outfalls to their corresponding sewersheds. Some outfalls are associated with multiple sewersheds, some sewersheds are associated with multiple outfalls, and some sewersheds are associated with the same relief sewer outfall. For these reasons, sewersheds were grouped based on interconnection. Some sewersheds were grouped based on common relief sewer outfalls and the overflow volume total for all of the outfalls in the grouping was used. A listing of the sewershed groupings can be seen here.

Comparing land use and land cover within each sewershed can help to explain differences in performance between different sewersheds. However, for statistical modeling purposes, comparing performance to each sewersheds’ unique land cover and land use characteristics reduces the chance of statistically significant results given the relatively few datapoints that exist. To solve this, a k-means clustering technique was used to place sewersheds in groups based on similar land use and land cover characteristics. The result is that every sewershed is placed into one of three land cover categories and one of five land use categories. For example, lower-density residential neighborhoods in the Northeast are held distinct from higher-density residential neighborhoods in South Philadelphia.

Constructed Projects

Project-specific data was obtained for all constructed project within the CSS boundary. This includes all public streets, public parcel, private incentives, and private mandate projects. All stormwater management projects were joined to the precipitation and CSO overflow data for following year (i.e. projects constructed in 2016 were joined to precipitation data from 2017). This is critical for the later analysis of the data so that all precipitation data is being analyzed from year following project completion in order to observe project effects on the CSS.

Public Projects

Project-specific data was obtained directly from the 2018 NPDES annual report and formatted. The location data for public streets projects and public point projects were obtained from OpenDataPhilly.com. The project-specific data was joined to the location data via Project ID and Work Number.

All streets projects were split proportionally if they traversed more than 1 sewershed. The total length falling within each drainage area was multiplied by each project’s total cost, the total number of the various types of SMPs, and design capacity. This gives a more accurate estimate of each project’s proportional effects if it falls in multiple sewer drainage areas.

Private Projects

Project-specific data was obtained directly from the 2018 NPDES annual report and formatted. Location data for private mandate projects and private incentives projects were obtained from OpenDataPhilly.com and joined to the project specific data via project tracking numbers.

Crosswalk and Join All Public and Private Projects

Since private projects and public projects are reported as different SMP types, a basic crosswalk was developed in order to coerce both datasets into a unified dataset. (Page 149 of 2018 NPDES Report) Being able to compare private project SMPs to public project SMPs is critical for later analysis of project types and SMP types and their relative effects on overflow.

New SMP Categories Private Project SMP’s Public Project SMP’s
GREEN_ROOF GREEN_ROOF SMP_GREENROOF
POROUS_PAVEMENT POROUS_PAVEMENT SMP_PERVIOUSPAVING
SUBSURFACE_BASIN SUBSURFACE_INFILTRATION_BASIN SUBSURFACE_DETENTION_BASIN & CISTERN
GREEN_SUBSURFACE_BASIN (none) SMP_BUMPOUT & SMP_TREETRENCH & SMP_PLANTER
SURFACE_BASIN SURFACE_INFILTRATION_BASIN & SURFACE_DETENTION_BASIN SMP_BASIN
BIOINFILTRATION BIOINFILTRATION & BIORETENTION SMP_RAINGARDEN & SMP_GREENGUTTER & SMP_SWALE & SMP_WETLAND & SMP_STORMWATERTREE

Outlier Removal

CSOs P-03, P-04, P-05, C-23, S-33, D-08, and D-02 were reporting over 100% overflow of the total precipitation volume. It is assumed that there is some other hydrological situation in those drainage areas that is not accounted for here, these CSOs are therefore dropped from the dataset so as to not interfere with the statistical analysis.

Limitations of the Data

A sample of the final dataset can be viewed here. Given the technical challenges of measuring CSO outfall volume, compounded by the year-long aggregation period, the dependent variable of this study (overflow volume) has an unknown margin of error. However, since overflow volume is one of the benchmark metrics reported to NPDES, it is assumed that the error is relatively consistent from year to year. So this study instead uses the relative change in overflow volume instead of gross annual overflow volume. Additionally, sewer balancing and CSO conveyance storage capacity is not taken into account in this study. Therefore, a change in a sewershed’s overflow volume may be in part due to changes in system balancing. This study seeks to use green stormwater infrastructure projects to explain changes in overflow volume and, without data on changes in system conveyance storage capacity and system balancing, assumes system balancing to be exogenous.

Results

CSS System-wide Performance

The chart below shows the variation in total volume that overflowed from each sewershed grouping each year. This shows the wide variation in annual overflow volumes across the CSS. That certain sewersheds contribute far greater volumes than others means the relative impact of stormwater projects is different for different sewersheds. Hover your mouse over the chart to see more information on the data.

However, comparing sewersheds’ overflow percent is perhaps a better measure of changes in system performance since it takes into account some degree of variation in precipitation. The chart below shows the percent of total annual rainfall volume that fell on each sewershed grouping that ended up overflowing, or percent overflow. Highlight areas on the chart to zoom in and hover your mouse over lines to identify values.

System-wide Performance Measures

Click on the tabs below and toggle between the charts to see annual trends aggregated for the entire CSS. Given the error associated with this level of aggregation, these trends are not significant enough to draw any conclusions. However, these charts show the relationship between precipitation and overflows as well as how the systems has reacted to variations in precipitation from year to year.

Total Volume

The chart below shows the variation in the total volume overflow (acre-inches).

Total Events

The chart below shows the variation in the total rain events and overflow frequency.

Total Heavy Events

The chart below shows the variation in the total heavy rain events and overflow frequency. This study considers a heavy rain event to be those that are greater than 1.5-inches over a 24-hour period.

The high degree of variation in annual precipitation volume and frequency makes it difficult to draw any system-wide conclusions about trends in CSO overflows over the past 10 years. Given the steady downward trend on rainfall in the last 5 years of the data (2013-2017), this time period was selected for further testing. This subset has a variation in rainfall volumes and total events with less outlier noise as seen in the total 10-year dataset. This subset allows for better linear regression analysis and included more SMPs than earlier years.

In order to determine whether the variation in annual CSO overflow volume is significantly different than that of annual rainfall, linear regressions were run for the volume intensity of both annual precipitation and annual CSO overflows. Volume intensity is the annual total volume divided by the total annual rain events. This measure accounts for years that had fewer but more intense rainfalls. The graph below shows that the volume and frequency of CSO overflows decreased faster (slope of -711) than that of precipitation (slope of -19), which indicates a reduction in the overall CSO system overflows. However, neither of these regressions is statistically significant, so further analysis is necessary in order to reject the null hypothesis that there has been no change in overflows.

Dependent variable:
volume_intensity precip_intensity
(1) (2)
Year -710.745 -18.715
(419.740) (277.182)
Constant 1,442,364.000 53,249.210
(845,775.700) (558,522.700)
Observations 5 5
R2 0.489 0.002
Adjusted R2 0.318 -0.331
Residual Std. Error (df = 3) 1,327.333 876.528
F Statistic (df = 1; 3) 2.867 0.005
Note: p<0.1; p<0.05; p<0.01



A linear model of the percent overflow compared to a linear model of overflow volume intensity as a function of precipitation volume intensity (overflow volume intensity / precipitation volume intensity) shows that volume and frequency have been reduced both independently of precipitation and as a function of precipitation. The slope of the percent overflow is -0.023 and is statistically significant. The slope of volume-intensity is -0.045 and is also statistically significant. These results indicate that both the volume and frequency of overflows have been reduced by roughly 2.3% over the 5 year period.

Dependent variable:
volume/TOTAL_PRECIP volume_intensity/precip_intensity
(1) (2)
as.integer(Year) -0.023** -0.045*
(0.005) (0.018)
Constant 0.447*** 91.898*
(0.038) (36.755)
Observations 5 5
R2 0.867 0.673
Adjusted R2 0.823 0.563
Residual Std. Error (df = 3) 0.017 0.058
F Statistic (df = 1; 3) 19.548** 6.162*
Note: p<0.1; p<0.05; p<0.01



Panel linear models with overflow percent and gross volume overflow as the dependent variables show a statistically significant trend that the increase in stormwater infrastructure is associated with a slight reduction in overflow percent. The second model below shows that each additional greened acre of stormwater infrastructure is associated with a reduction in overflow volume by roughly 107 acre-inches annually. The 3rd model below also indicates with statistical significance that every additional greened acre of public stormwater infrastructure (streets and parcel projects) is associated with roughly 120 acre-inches of annual overflow reduction. These results indicate correlation between infrastructure and overflow reduction. Further analysis at a more detailed scale will start to parse apart causation.

Dependent variable:
overflow_percent volume
All Projects All Projects Public Projects Private Projects
(1) (2) (3) (4)
TOTAL_GA -0.003*** -107.328*** -119.879*** 7.687
(0.0005) (16.039) (38.403) (14.159)
TOTAL_PRECIP 0.872*** 0.871*** 0.745***
(0.018) (0.022) (0.018)
Observations 780 780 491 601
R2 0.049 0.891 0.801 0.793
Adjusted R2 -0.142 0.868 0.756 0.742
F Statistic 33.426*** (df = 1; 649) 2,637.163*** (df = 2; 648) 802.651*** (df = 2; 399) 923.314*** (df = 2; 481)
Note: p<0.1; p<0.05; p<0.01

Individual Sewershed Performance

Disagregating and analyzing each sewershed separately gives a much more nuanced understanding of factors associated with overflow reduction. The animation below shows that the disagregating data is much more complicated and does not readily lend itself to easy interpretation. Notice how the addition of stormwater infrastructure does not always appear to lower the overflow percent for some sewersheds. Also notice that the percent imperviousness does not appear to be correlated with any trends in percent overflow.

In order to examine which sewersheds have decreased volume overflow at faster rates, bivariate regressions were taken for each sewershed to find the rate of change in the percent overflow over the 5 year period. To better understand this process, notice how the regression lines were taken for the four sample sewersheds below. Sewersheds with a steeper negative slope have reduced percent overflow at a higher rate than those of lower slopes.

This same process was applied to all sewersheds in the CSS area, the result is the chart below. Highlight portions of the chart to zoom in and hover your mouse over the data for more information.

Plotting the rate change in overflow percent against the total greened acres within each sewershed shows a slightly negative correlation between the amount of greened acres and the percent overflow. In other words, the more greened acres a sewershed has, generally the lower it’s volume overflow over the 5 year period. However, this downward trend is not statistically significant and is explained more by the total heavy rain events than by stormwater infrastructure investment. This plot also shows that green infrastructure has been more successful at reducing CSO overflows in some sewersheds than in others.

Dependent variable:
rate_chng
TOTAL_GA -0.00001
(0.0001)
TOTAL_HEAVY_EVENTS -0.006***
(0.001)
Constant 0.316***
(0.066)
Observations 108
R2 0.217
Adjusted R2 0.202
Residual Std. Error 0.031 (df = 105)
F Statistic 14.556*** (df = 2; 105)
Note: p<0.1; p<0.05; p<0.01



Part of why there is not a significant trend in rate change and greened acres may be due to factors other than impervious land cover and greened acres, but rather due to sewer system storage capacity and balancing. The plot below shows, unsurprisingly, a strong linear relationship between the volume of precipitation and the volume of CSO overflow. However, what is surprising about this plot is that there is no apparent correlation between impervious land cover and percent overflow. This plot and the linear model indicate that a seweshed’s percent impervious land cover is not correlated with percent overflow. Project prioritization may need more nuanced metrics than sewershed impervious cover. Land cover mix and land use mix may help to explain this variation.

Dependent variable:
overflow_percent
impervious_pct 0.085
(0.062)
Constant 0.139***
(0.046)
Observations 173
R2 0.011
Adjusted R2 0.005
Residual Std. Error 0.099 (df = 171)
F Statistic 1.874 (df = 1; 171)
Note: p<0.1; p<0.05; p<0.01



In order to determine actionable insight on which sewersheds have performed better and why, each sewershed’s rate change was further analyzed. Taking a look at the density distribution of the rate changes in percent overflow reveals a bimodal distribution for sewersheds with green infrastructure. There is evidence that an additional variable is at play which is causing one group of sewersheds to lower overflow percent while the other group seems to remain unchanged by green infrastructure. The bimodal distribution indicates a separation between sewersheds where green infrastructure has performed well and where it has not performed significantly different than sewersheds with no GSI interventions. The sewersheds with no GSI serve as a control group with which to compare the magnitude change in sewersheds with GSI interventions.

Splitting the GSI data at the parabolic vertex reveals 2 normally distributed data sets, one where GSI has reduced volume percent and the other GSI has not caused any change in overflow about the control group.

Land Use

Click here to see the characteristics of each land use cluster. When examining the rate change by the land use clusters, there is a separation in performance based on the mix of land use within each sewershed. Clusters 3, 4, and 5 appear to to be linked with better performance than clusters 1 and 2, which have generally shown no change above the control group. Clusters 1 and 2 are prominently residential land uses, which indicates that those land uses are under-performing. This is perhaps because residential properties and parcels under 15,000 square feet are not billed based on impervious cover and not eligible for SMIP grants.

The bar plot below shows that cluster 5 has seen significantly greater reductions in overflow percent due to stormwater management infrastructure than the other land use clusters. Cluster 5 are areas that are high in institutional land uses, which may indicate the success of the Stormwater Management Incentives Program at funding high-impact retrofits on institutional properties.

The below linear model shows that sewersheds with higher percentages of industrial, office/mixed use, civic/cultural/institutional, retail and vacant land uses have reacted the best to stormwater infrastructure in reducing percent overflow. This may be due to a higher proportional investment in these land uses.

Dependent variable:
overflow_percent
officeMixed_pct 0.742***
(0.193)
civicCultural_pct 0.758***
(0.182)
openSpace_pct 0.175
(0.199)
Residential_pct 0.718***
(0.182)
Retail_pct 0.751***
(0.185)
Industrial_pct 0.494***
(0.182)
Transportation_pct 0.637***
(0.180)
Vacant_pct 0.959***
(0.204)
Constant -0.460**
(0.179)
Observations 1,169
R2 0.062
Adjusted R2 0.056
Residual Std. Error 0.127 (df = 1160)
F Statistic 9.636*** (df = 8; 1160)
Note: p<0.1; p<0.05; p<0.01

Land Cover

The same analysis was done for land cover clusters as explained above. Click here to see the characteristics of each land cover cluster. It is clear that clusters 3 and 3 have performed better than than cluster 1. The bimodal distribution of all three clusters indicate that there is an exogenous factor that the clustering does not explain. Of the land cover clusters, cluster 2 reacted the least to stormwater infrastructure and saw nearly no change from the control group. This may indicate a variation in types of stormwater infrastructure across sewersheds within the same cluster. Further investigation into whether the cluster 2 sewersheds that did not react to infrastructure consisted of a certain SMP or program type is necessary.

The linear model below shows that the percent building cover correlated to roughly 40% higher overflow percent than roads. The R2 of this model is lower and there are fewer significant independent variables than the land use model. This is important because it shows that land cover is not as good at explaining overflow reduction as land use. It is also important to note that street density does not increase percent overflow as significantly as building density. This may have implications for project prioritization strategy.

Dependent variable:
overflow_percent
pavement_pct 0.180
(0.199)
roads_pct 0.338*
(0.196)
buildings_pct 0.549***
(0.191)
tree_pct 0.083
(0.224)
grass_pct 0.270***
(0.096)
Constant -0.151
(0.193)
Observations 1,169
R2 0.042
Adjusted R2 0.038
Residual Std. Error 0.129 (df = 1163)
F Statistic 10.245*** (df = 5; 1163)
Note: p<0.1; p<0.05; p<0.01

Program Type

To better understand what additional factors have led to some investments performing better than others, the data is plotted by the percent of program type for each sewershed. This plot represents an average of the percent of total GA in each watershed that is under each program type. The chart shows that sewersheds with higher rates of regulation green infrastructure tend to not see a reduction in overflow percent. This is likely because the loading ratios required by regulation are typically far lower than the loading ratios of public projects. The chart also indicates that sewersheds that have not seen overflow volume reduction tend to have more streets projects as a share of total greened acres.

The panel linear model below shows the relative effectiveness of each program type in reducing volume overflow. The results show that the most effective program has been public parcel projects. The data show that there is a statistically significant correlation between lower overflows and streets and public parcel programs. However, there are around 4 times more streets project greened acres than public projects, which may be the reason for the streets coefficient being stronger than public projects. Regulation projects are correlated with higher overflow percents. This is likely due to the fact that the loading ratios and design capacities of regulation infrastructure is typically much lower than those of public streets and parcel projects.

Dependent variable:
overflow_percent volume
Overflow Percent Overflow Volume
(1) (2)
RETROFIT_GA -0.001* 4.777
(0.001) (16.529)
REGULATION_GA -0.001 -45.770
(0.002) (46.951)
STREETS_GA -0.010*** -247.203***
(0.004) (74.775)
PUBLIC_PARCEL_GA -0.003 -1,233.387***
(0.004) (75.974)
TOTAL_HEAVY_EVENTS 0.001**
(0.001)
TOTAL_PRECIP 0.813***
(0.018)
Observations 780 780
R2 0.067 0.921
Adjusted R2 -0.127 0.905
F Statistic (df = 5; 645) 9.201*** 1,506.285***
Note: p<0.1; p<0.05; p<0.01



Relative Investment

The plots below compare relative greened acres investment to overflow volume by program type for each sewershed. Sewersheds that fall below the line have more than average stormwater infrastructure relative to overflow volume, and are probably over-invested. The sewersheds that fall above the line have lower than average stormwater infrastructure investment relative to volume overflow, and are under-invested relative to other sewersheds as a function of overflow volume.

All Program Types

The plot below shows sewersheds the trend in total infrastructure and total volume overflow.

Public Parcel Projects

The same chart is plotted against public parcel projects. In this chart, sewersheds that fall above the line are under-invested with public parcel projects.

#### Public Streets Projects

The same chart is plotted against public streets projects. In this chart, sewersheds that fall above the line are under-invested with streets projects and those falling below the line are over-invested relative to other sewersheds.

Private Incentives Projects

The chart is plotted against private incentives projects. In this chart, sewersheds that fall above the line are under-invested with incentives projects and those falling below the line are over-invested relative to other sewersheds.

Private Mandate Projects

Lastly, the same chart is plotted against private regulation projects. In this chart, sewersheds that fall above the line are not receiving redevelopment regulations projects as the same rate as sewersheds below the line. Those sewersheds falling above the line should be prioritized for public streets and public parcel projects to help fill the void left by the lack of redevelopment regulation stormwater projects. Those sewersheds falling below the line should receive lower priority for public projects and incentives if the decision-making process is entirely system performance driven.

Investment Parities by Sewershed

The table below shows relative scaled residual values from the above linear models. The scaled residual values, in this case, represent a relative unit of comparison between sewersheds. For example, if a sewershed is listed as having a Private Mandate of -1, then that sewershed is perhaps the most under-invested sewershed in Private Mandates relative to volume overflow and relative to Private Mandates in other sewersheds. conversely, values closer to 1 indicate sewersheds that are further along in certain program type investments than others. This is a performance-based assessment that, of course, excludes many factors that are highly important in infrastructure planning and merely offers a way to benchmark sewersheds against each other.



SMP Type

The chart below shows the relative use of each SMP type category in both sewersheds that saw a reduction in overflow volume versus those sewersheds that did not see a relative improvement. Sewersheds that saw a higher decrease in overflow percent tended to have higher shares of subsurface and green subsurface basins. Additionally, bioinfiltration SMPs had a higher association with sewersheds with decreased overflows than were surface basins.


The panel linear model below shows the relative effectiveness of various SMP types across the dataset. Porous pavement, green subsurface basins, and surface basins are correlated with reductions in total volume whereas greenroofs tend to be associated with an increase in total overflow volume. This may be due to a preference for greenroofs in private mandate projects. This is not due to individual SMP type performance, but rather a result of the amount of each SMP type in each sewershed. This may indicate that porous pavement, green subsurface basins, and surface basins have been more effectively located in waterhseds that have seen large reductions.

Dependent variable:
volume
GREEN_ROOF 680.080***
(141.162)
POROUS_PAVEMENT -565.920***
(127.935)
SUBSURFACE_BASIN 78.502
(83.003)
GREEN_SUBSURFACE_BASIN -214.330***
(50.532)
SURFACE_BASIN -1,303.983***
(199.842)
BIOINFILTRATION 88.297
(58.765)
TOTAL_PRECIP 0.850***
(0.021)
Observations 780
R2 0.899
Adjusted R2 0.877
F Statistic 815.231*** (df = 7; 643)
Note: p<0.1; p<0.05; p<0.01



Discussion

The effects of the green stormwater program are slight but statistically significant. Philadelphia’s stormwater infrastructure program is measurably reducing the overall CSO overflow volume and frequency. However, it is evident that stormwater infrastructure has worked better in some sewersheds than others. Part of the parity in performance has to do with market patterns of redevelopment in the city, but it also appears that the infrastructure investment decision-making process could potentially be more heavily guided by CSO overflow performance. For example, the CSO regulators that are overflowing the highest volumes seem to be under-invested both in public projects as well as private incentives in terms of magnitude of overflow. Perhaps adopting a prioritization process which weights prospective projects by the sewershed in which they reside and identifies the most effective program types for each sewershed could help guide the next phase of the Green Cities Clean Waters program. Sewersheds with higher rates of redevelopment might receive less investment in public projects, while sewersheds with higher overflow volumes might have a higher cost-per-green-acre investment threshold.

It is apparent from this study that the most effective performance-based approach in many of the watersheds is to build large, high-capacity public parcel projects which utilize some mixture of surface detention, subsurface detention, and green subsurface detention. For sewersheds where parcel projects have shown improvements, a site prioritization based on existing hydrology to identify open space parcels for larger-scale public parcel projects may offer a data-driven process for site identification.

Appendix

Data Wrangling

Precipitation Modeling

Raw Daily Precipitation Values by Year (2018)

Annual Totals By Region

Total Precipitation

Total Heavy Precipitation Events

A threshold of 1.5" was set to determine the dummy variable for heavy rain events. This threshold was chosen because 1.5" was the initial design storm event for private regulations green infrastructure in Philadelphia.

Total Precipitation Events

Interpolating Precipitation Raster Surfaces

Inverse distance weighting is used to interpolate a raster surface across the region that averages together the variations in rain data from the different weather stations.

Total Annual Precipitation

Total Annual Rain Events

Total Heavy Precipitation Events

CSO Drainage Areas

The sewershed boundaries (CSO drainage areas) shapefile was obtained from PWD and is the only data input for this study which is not publicly available. Note that the sewershed boundaries are not presented in this report due to the sensitive nature of that data. The map below instead presents watershed boundaries that were calculated in ArcHydro to serve as a visual proxy for the sewershed boundaries. The sewershed boundaries did not initially match the CSO regulators listed in the NPDES documentation, so a thorough examination of all NPDES annual reports from 2009-2018 guided the process of matching sewersheds to CSO regulators. In some cases relief sewers with new CSO regulators were introduced, in other case relief sewers were consolidated into other regulators.

Sewer drainage areas are grouped according to downstream outfall points. Some sewer drainage areas are grouped by downstream relief sewer overflow. These were grouped according to documentation in PWD’s annual NPDES reports as well as historical sewer maps published on PhillyWatersheds.org. CSO overflow volume data and locations were obtained directly from PWD’s yearly NPDES reports from 2009-2018. The data was formatted in order to be unified across all 10 years.

CSO Groupings

Qoutfall Grouping
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG
C01
C02
C04
C05
C06
C07
C09
C10-C11-R24-CFRA
C12
C13
C14
C15
C16
C17-R01-CFRTR
C18
C19
C20
C21
C22
C25
C28A
C29
C30
C31
C32
C33
C34
C35
C36
C37
D03
D04
D05-D07
D06
D09
D11
D12
D13
D15
D17-D19-D20-D21-D22
D18
D23
D25-D39-D44-D45-D47-R07-R12-R12R-R19-S05-SFRM
D37-D38
D40
D41
D42
D43
D46
D48
D49
D50
D51
D52
D53
D54
D58
D61
D62
D63
D64
D65
D66
D67
D68-S42-S42A
D69
D70
D71
D72-D73
F03
F04
F05
F06-F09
F07
F08
F10
F11
F12
F13
F21-R13-R14-DFRW
F23
F24
F25
P01
P02
S01
S02
S03
S04
S06-S07
S08
S09
S10
S11
S13
S14
S15
S16
S17
S18
S19
S20
S21
S22
S23
S24
S25
S26
S27-S50-S51
S30
S31
S32
S36
S37
S38
S44
S45
S46
T01
T03
T04
T05
T06
T07
T08-R15-TFRRR
T09
T10
T11
T12
T13-T14-T15-R18-FFRFG

Mapping Stormwater Projects

Clustering Sewersheds

The purpose of clustering sewersheds is to improve the statistical modeling process by summarizing land use and land cover variations into fewer data points. The clustering allows models to compare like sewersheds to like sewersheds in order to observe significant differences in dependent variables. The clustering also allows for aggregation of sewershed data into other like sewersheds, which is useful for smaller sewersheds which may contain few associated data points. These clusters could also be use in the project planning process in determining successful or more cost effective program types and SMP types for each sewershed.

Clustering by Land Cover

K-means clustering analysis is performed on a 5-dimensional dataset of land cover. The land cover data was processed from the 2008 Philadelphia Land Cover raster via the OpenDataPhilly.com API. The resulting clusters will put all of the drainage areas into groups according to similar land cover mixes. The number of clusters is decided by an elbow plot by selecting the number of clusters at the inflection point of the plot.

The chart below shows clear distinctions between the various clusters which are representative of the land cover mixes around Philadelphia.

Clustering by Land Use

K-means clustering analysis is performed on a 8-dimensional dataset of use percentages. The land use data was processed from the City of Philadelphia Land Use Map via the OpenDataPhilly.com API. The resulting clusters will put all of the drainage areas into groups according to similar land use mixes. The number of clusters is decided by an elbow plot by selecting the number of clusters at the inflection point of the plot.

The chart below shows clear distinctions between the various clusters which are representative of the land use mixes around Philadelphia.{#cluster_LU}

Outlier Removal

Before Removal

After Removal

The Dataset

This dataset represents the first 20 rows of the data used in the study.
OUTFALL acres AVERAGE_PRECIP TOTAL_PRECIP TOTAL_HEAVY_EVENTS TOTAL_EVENTS COST GREEN_ROOF POROUS_PAVEMENT SUBSURFACE_BASIN GREEN_SUBSURFACE_BASIN SURFACE_BASIN BIOINFILTRATION TREES RETROFIT_GA REGULATION_GA STREETS_GA SCHOOLS_GA PARKING_GA VACANT_GA OPEN_SPACE_GA TOTAL_GA Year Frequency duration volume Residential_lowDensity_pct Residential_medDensity_pct Residential_hiDensity_pct Retail_pct Office_pct Mixed_Use_pct Industrial_pct Civic_Institution_pct Transportation_pct Culture_Recreation_pct Park_openSpace_pct Cemetary_pct Vacant_pct pavement_pct roads_pct buildings_pct tree_pct grass_pct impervious_pct overflow_percent landCover_cluster landUse_cluster PUBLIC_PARCEL_GA
C01 28.17910 42.54630 1198.9165 67.78168 127.6801 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 21 25.25 102.265317 0.0017719 0.3169780 0.0059902 0.2648861 0.0448340 0.0120936 0.0179940 0.0383764 0.2874653 0.0096245 0.0000000 0.0000000 0.0000000 0.4222116 0.1659512 0.2445479 0.0530973 0.2286095 0.8327107 0.0852981 1 3 0
C02 4.25668 42.61728 181.4081 67.90693 127.8416 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 8 6.25 9.959862 0.5297611 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.2270404 0.0000000 0.0040987 0.0000000 0.0000000 0.0744522 0.0568706 0.1076361 0.3447398 0.3525599 0.2389589 0.0549031 2 4 0
C04 114.88350 42.76935 4911.2590 68.24816 128.3084 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 27 61.50 1248.489783 0.5284032 0.1157077 0.0206842 0.0000000 0.0000000 0.0009488 0.0000000 0.0059134 0.2357871 0.0000000 0.0624524 0.0000000 0.0100037 0.1032443 0.0947193 0.1551628 0.3854448 0.4820570 0.3531263 0.2542097 2 4 0
C05 25.24770 42.67308 1077.3972 67.99041 127.9639 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 20 27.00 142.561460 0.0000000 0.5298449 0.1360700 0.0099510 0.0000000 0.0189787 0.0000000 0.0032143 0.2850262 0.0167061 0.0002190 0.0000000 0.0000000 0.2953301 0.1547387 0.2589531 0.1089308 0.3644187 0.7090218 0.1323202 1 4 0
C06 102.18700 42.88831 4382.6273 68.51712 128.6312 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 68 227.25 2006.047675 0.0948542 0.4853622 0.0019670 0.0040041 0.0005409 0.0091059 0.0000000 0.0458944 0.3483550 0.0000350 0.0000133 0.0000000 0.0098804 0.2386553 0.1324491 0.3061754 0.1676813 0.3102960 0.6772797 0.4577272 1 4 0
C07 43.19460 42.86258 1851.4321 68.45083 128.4905 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 32 69.25 569.297372 0.0484127 0.3419775 0.0036348 0.0723910 0.0000000 0.0225534 0.0017425 0.1262312 0.3217164 0.0302802 0.0188094 0.0000000 0.0122624 0.2756239 0.1756939 0.2943594 0.0940046 0.3204432 0.7456772 0.3074903 1 3 0
C09 71.78380 42.96261 3084.0195 68.62353 128.7220 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 42 83.00 607.448851 0.0814248 0.3666352 0.0195930 0.0216989 0.0013271 0.0224155 0.0131171 0.0452022 0.3185459 0.0700471 0.0198262 0.0000000 0.0201781 0.2165018 0.1563349 0.3197896 0.0732930 0.4680907 0.6926264 0.1969666 2 4 0
C10-C11-R24-CFRA 358.09697 43.03482 15422.9889 68.78844 128.9044 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 29 83.75 5082.002369 0.0807920 0.4265335 0.0144616 0.0153557 0.0070963 0.0335090 0.0174051 0.0227469 0.3572915 0.0015884 0.0002778 0.0000000 0.0229522 0.1808645 0.1769023 0.3705775 0.0933710 0.3568631 0.7283443 0.3295083 1 4 0
C12 56.07680 42.92825 2407.2789 68.52123 128.5359 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 50 150.00 761.842231 0.0059134 0.4437824 0.0251490 0.0057991 0.0000000 0.0074623 0.0154781 0.0652808 0.3831331 0.0316302 0.0110281 0.0000000 0.0053511 0.1829345 0.1739285 0.3760601 0.1186664 0.2964877 0.7329230 0.3164744 3 4 0
C13 49.10260 42.77332 2100.2813 68.12591 128.0637 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 42 96.50 477.763912 0.4494880 0.1395136 0.0019250 0.0004965 0.0007156 0.0020721 0.0056381 0.0068860 0.3591217 0.0050671 0.0253585 0.0000000 0.0037221 0.1463442 0.1788739 0.3352798 0.1331356 0.4122999 0.6604979 0.2274761 2 4 0
C14 76.69730 42.79051 3281.9163 68.14466 128.0883 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 42 132.25 1212.156187 0.0818838 0.3721675 0.0045012 0.0502107 0.0000000 0.0071716 0.0026718 0.1060953 0.3551885 0.0000000 0.0092842 0.0000000 0.0108299 0.2073983 0.1638132 0.3564937 0.1168978 0.3108532 0.7277051 0.3693440 3 4 0
C15 10.39510 42.82407 445.1604 68.25436 128.2433 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 27 50.50 152.096118 0.0056680 0.3734172 0.0398995 0.0207941 0.0000000 0.0100129 0.0118304 0.0916341 0.3897237 0.0000000 0.0229594 0.0000000 0.0342801 0.2181994 0.2133077 0.3053595 0.0788542 0.3677653 0.7368667 0.3416658 1 3 0
C16 4.87572 42.81443 208.7512 68.22381 128.1369 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 7 6.50 16.765510 0.0826675 0.0177893 0.4701987 0.0000000 0.0000000 0.0000000 0.0000000 0.0963913 0.3299780 0.0000000 0.0000000 0.0000000 0.0029793 0.0539300 0.1533761 0.1502827 0.4157047 0.4519123 0.3575889 0.0803134 2 4 0
C17-R01-CFRTR 1075.99500 43.19187 46475.7853 69.09236 129.1325 0 1 3 5 0 0 0 0 0 2.2 0 0 0 0 0 2.2 2009 71 462.50 25715.465297 0.0458911 0.4155321 0.0153606 0.0246225 0.0034827 0.0221830 0.0150410 0.0605969 0.3628356 0.0053364 0.0025269 0.0015188 0.0250799 0.2011067 0.1781144 0.3892641 0.0881522 0.2870879 0.7684851 0.5533089 3 4 0
C18 98.94820 42.78225 4233.2269 68.04450 127.9763 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 39 110.25 1270.779210 0.0238418 0.3911858 0.0319032 0.0143120 0.0060548 0.0227518 0.0000000 0.0311508 0.3545586 0.0713809 0.0008145 0.0120098 0.0400369 0.1823586 0.1833228 0.3524940 0.0922137 0.3793347 0.7181754 0.3001916 1 4 0
C19 64.74420 42.41682 2746.2434 67.25861 127.0381 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 21 23.00 289.024738 0.0388644 0.2425466 0.0000000 0.0264146 0.0000000 0.0050340 0.0404877 0.0048817 0.3468270 0.0000000 0.0013490 0.2220262 0.0715698 0.1439539 0.1570144 0.2075279 0.1053989 0.7724401 0.5084961 0.1052437 2 3 0
C20 21.03050 42.22930 888.1032 66.75263 126.5138 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 19 25.50 153.015328 0.0843598 0.5189594 0.0141999 0.0050729 0.0020001 0.0068940 0.0000000 0.0185333 0.3327183 0.0000000 0.0000000 0.0113345 0.0059287 0.2730249 0.1637909 0.3271627 0.0745627 0.3228203 0.7639785 0.1722945 1 4 0
C21 29.43910 42.09135 1239.1313 66.48021 126.2434 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 20 28.75 200.940577 0.0400714 0.3182770 0.0039836 0.0000000 0.0000000 0.0002954 0.0000000 0.1306525 0.3398331 0.0000000 0.1616229 0.0000000 0.0052602 0.2844224 0.1774164 0.2306812 0.0675087 0.4799343 0.6925201 0.1621625 1 3 0
C22 59.19700 41.96869 2484.4206 66.10531 125.8624 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0.0 2009 40 83.50 668.690928 0.0712054 0.2531666 0.0000000 0.0346941 0.0069887 0.0017726 0.1687710 0.0147334 0.3962964 0.0000000 0.0006450 0.0000000 0.0517240 0.2211469 0.1862693 0.2413305 0.1370228 0.4284918 0.6487467 0.2691537 2 3 0
C25 44.59510 41.82736 1865.2952 65.79782 125.4565 0 0 0 1 0 0 0 0 0 0.8 0 0 0 0 0 0.8 2009 26 70.00 980.558347 0.0370066 0.2131961 0.0660626 0.0634235 0.0017332 0.0266448 0.0448083 0.1648131 0.3280780 0.0000000 0.0000000 0.0000000 0.0542307 0.2734944 0.1781523 0.2612704 0.1208656 0.3326987 0.7129171 0.5256853 1 3 0

Site Selection Aproach

The T-14 sewershed is, by a high margin, the largest volume CSO overflow in Philadelphia, yet it has received lower than average redevelopment and total greened acre investment.This makes it a test region for a proposed data-driven site selection process for large-scale public parcel projects. The map below shows a proof-of-concept for a hydrological modeling approach to site prioritization. The digital elevation model of the sewershed is processed into a flow accumulation raster in order to identify natural flow paths of surface runoff. The areas of higher flow accumulation, in theory, are better candidates for a local area disconnection public parcel project since areas of high flow accumulation indicate consolidation of drainage areas and the required separated sewer conveyance would likely mirror topography. All vacant and public open space parcels are selected and inherit the maximum flow accumulation value of the raster surface in the area of the parcel. This identifies parcels that might be good candidates because some part of the parcel lies in an area of high flow accumulation.